OpenAI Symphony

Set up and run OpenAI Symphony with isolated issue workspaces, workflow contracts, and unattended Codex orchestration for Linear projects.

This skill appears to do what it claims, but it operates at a high trust level. Before installing or enabling unattended runs: 1) only use a test Linear project and a non-production repository until you validate one end-to-end run; 2) minimize token scopes — prefer a least-privilege Git token or deploy key rather than a broad GITHUB_TOKEN; 3) prefer SSH key auth if you don't want to store a token; 4) verify the 'codex' binary is legitimate and understand whether it uses OPENAI_API_KEY or a separate login session; 5) inspect any repository hooks (after_create, before_run, etc.) you will run — hooks can execute arbitrary code from the repo; 6) keep workspace root on a dedicated path (not a parent/shared directory) and do not store secrets in ~/symphony/ memory files; 7) require explicit user approval before enabling unattended/autonomous operation; and 8) if provenance matters, confirm the referenced upstream implementation (e.g., the GitHub repo cited in docs) before trusting the skill. If you want, provide the exact token scopes you plan to use and the location of the codex binary and I can give more specific hardening steps.

Type: OpenClaw Skill Name: symphony Version: 1.0.0 The OpenAI Symphony skill provides unattended orchestration for Linear issues and Git repositories, requiring high-privilege credentials (LINEAR_API_KEY, GITHUB_TOKEN) and executing shell hooks (SKILL.md, WORKFLOW.md). While the skill includes comprehensive safety documentation and isolation policies (safety-guardrails.md), the inherent risks of shell execution and broad API access for its stated purpose align with the suspicious classification. No evidence of intentional malice or data exfiltration was identified.